predict only diff's, and allow diffs of arbitrary length

[jaredgreen2]

the diffs i am getting are being cut off and rejected for being too long, and shortening the file seems to solve the issue. im not sure why that is, and i may not always be able to do so. what is the default limit on the length of diffs? how do I change it? why would the length of the file(before the proposed edit) matter if the diff itself is relatively small?

[dosubot[bot]]

Hi @jaredgreen2! I'm Dosu and I’m helping the Roo-Code team.

I couldn't find any documentation specifying a default limit on the length of diffs in Roo Code or a direct way to change such a limit. However, there are some relevant settings and behaviors that may explain what you're seeing:

• Roo Code's diff application uses fuzzy matching with configurable confidence thresholds and a context window (default 40 lines) for matching changes. Performance and reliability can degrade with very large files or highly repetitive code, which can cause diffs to be cut off or rejected, even if the diff itself is small. This is because the system may struggle to confidently locate the correct context for the diff in a large file, or may hit internal resource limits during processing. "apply_diff" tool details
• The "Enable editing through diffs" setting is on by default and is designed to prevent truncated writes by rejecting incomplete file edits, which can happen with large files or complex instructions. Disabling this setting will cause Roo Code to rewrite the entire file for every edit, which is slower and uses more tokens. "Fast Edits" feature
• Roo Code improved detection of incomplete file writes by considering the predicted file length, so the length of the original file can directly impact whether a diff is accepted or rejected, even if the diff itself is small. "v2.2.21 release notes"
• The "File read auto-truncate threshold" setting controls how many lines are read from a file in one batch. Lowering this value can help with performance on very large files, but may require more read operations. "Tips and Tricks"

If you frequently hit these issues, try reducing the file size, increasing the uniqueness of the context around your diffs, or adjusting the fuzzy diff matching threshold in settings. Unfortunately, there is no documented way to set an explicit diff length limit or override it directly. If this limitation is blocking your workflow, you may want to monitor the documentation for updates or reach out to Roo Code support for more details.

To reply, just mention my name @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}   

[jaredgreen2]

can we make the models directly provide the line numbers where the diff segments begin?